Non-Gaussianity
as a Probe of the Physics of the Primordial Universe
Eiichiro Komatsu
(Texas Cosmology Center, University of Texas at Austin)
Solvary Workshop, “Cosmological Frontiers in Fundamental Physics”
May 13, 2009
1
How Do We Test Inflation?
•
How can we answer a simple question like this:•
“How were primordial fluctuations generated?”2
Power Spectrum
•
A very successful explanation (Mukhanov & Chibisov;Guth & Pi; Hawking; Starobinsky; Bardeen, Steinhardt &
Turner) is:
•
Primordial fluctuations were generated by quantum fluctuations of the scalar field that drove inflation.•
The prediction: a nearly scale-invariant power spectrum in the curvature perturbation, ζ:•
Pζ(k) = A/k4–ns ~ A/k3•
where ns~1 and A is a normalization. 3n s <1 Observed
•
The latest results from the WMAP 5-year data:•
ns=0.960 ± 0.013 (68%CL; for tensor modes = zero)•
ns=0.970 ± 0.015 (68%CL; for tensor modes ≠ zero)•
tensor-to-scalar ratio < 0.22 (95%CL)•
ns≠1: another line of evidence for inflation•
Detection of non-zero tensor modes is a next important stepKomatsu et al. (2009)
4
Anything Else?
•
One can also look for other signatures of inflation. For example:•
Isocurvature perturbations•
Proof of the existence of multiple fields•
Non-zero spatial curvature•
Evidence for “Landscape,” if curvature is negative.Rules out Landscape ideas if positive.
•
Scale-dependent ns (running index)•
Complex dynamics of inflation 5Anything Else?
•
One can also look for other signatures of inflation. For example:•
95%CL limits on Isocurvature perturbations•
S/(3ζ) <0.089 (axion CDM); <0.021 (curvaton CDM)•
95%CL limits on Non-zero spatial curvature•
Ω–1<0.018 (for Ω>1); 1–Ω<0.008 (for Ω<1)•
95%CL limits on Scale-dependent ns•
–0.068 < dns/dlnk < 0.0126
Komatsu et al. (2009)
positive curvature negative curvature
Beyond Power Spectrum
•
All of these are based upon fitting the observed power spectrum.•
Is there any information one can obtain, beyond the power spectrum?7
Bispectrum
•
Three-point function!•
Bζ(k1,k2,k3)= <ζk1ζk2ζk3> = (amplitude) x (2π)3δ(k1+k2+k3)b(k1,k2,k3)
8
model-dependent function
k1
k2
k3
Why Study Bispectrum?
•
It probes the interactions of fields - new piece of information that cannot be probed by the power spectrum•
But, above all, it provides us with a critical test of the simplest models of inflation: “are primordialfluctuations Gaussian, or non-Gaussian?”
•
Bispectrum vanishes for Gaussian fluctuations.•
Detection of the bispectrum = detection of non-Gaussian fluctuations 10
Gaussian?
WMAP511
Take One-point Distribution Function
•The one-point distribution of WMAP map looks pretty Gaussian.
–Left to right: Q (41GHz), V (61GHz), W (94GHz).
•Deviation from Gaussianity is small, if any.
12
Spergel et al. (2008)
Inflation Likes This Result
•
According to inflation (Mukhanov & Chibisov; Guth & Yi;Hawking; Starobinsky; Bardeen, Steinhardt & Turner), CMB anisotropy was created from quantum
fluctuations of a scalar field in Bunch-Davies vacuum during inflation
•
Successful inflation (with the expansion factor more than e60) demands the scalar field be almost interaction-free•
The wave function of free fields in the ground state is a Gaussian!13
But, Not Exactly Gaussian
•
Of course, there are always corrections to the simplest statement like this.•
For one, inflaton field does have interactions. They are simply weak – they are suppressed by the so-calledslow-roll parameter, ε~O(0.01), relative to the free-field action.
14
A Non-linear Correction to Temperature Anisotropy
•
The CMB temperature anisotropy, ΔT/T, is given by the curvature perturbation in the matter-dominated era, Φ.•
One large scales (the Sachs-Wolfe limit), ΔT/T=–Φ/3.•
Add a non-linear correction to Φ:•
Φ(x) = Φg(x) + fNL[Φg(x)]2 (Komatsu & Spergel 2001)•
fNL was predicted to be small (~0.01) for slow-roll models (Salopek & Bond 1990; Gangui et al. 1994)15
For the Schwarzschild metric, Φ=+GM/R.
f NL : Form of B ζ
•
Φ is related to the primordial curvature perturbation, ζ, as Φ=(3/5)ζ.•
ζ(x) = ζg(x) + (3/5)fNL[ζg(x)]2•
Bζ(k1,k2,k3)=(6/5)fNL x (2π)3δ(k1+k2+k3) x[Pζ(k1)Pζ(k2) + Pζ(k2)Pζ(k3) + Pζ(k3)Pζ(k1)]
16
f NL : Shape of Triangle
•
For a scale-invariant spectrum, Pζ(k)=A/k3,•
Bζ(k1,k2,k3)=(6A2/5)fNL x (2π)3δ(k1+k2+k3)x [1/(k1k2)3 + 1/(k2k3)3 + 1/(k3k1)3]
•
Let’s order ki such that k3≤k2≤k1. For a given k1, one finds the largest bispectrum when thesmallest k, i.e., k3, is very small.
•
Bζ(k1,k2,k3) peaks when k3 << k2~k1•
Therefore, the shape of fNL bispectrum is the squeezed triangle!(Babich et al. 2004) 17
B ζ in the Squeezed Limit
•
In the squeezed limit, the fNL bispectrum becomes:Bζ(k1,k2,k3) ≈ (12/5)fNL x (2π)3δ(k1+k2+k3) x Pζ(k1)Pζ(k3)
18
Single-field Theorem (Consistency Relation)
•
For ANY single-field models*, the bispectrum in the squeezed limit is given by•
Bζ(k1,k2,k3) ≈ (1–ns) x (2π)3δ(k1+k2+k3) x Pζ(k1)Pζ(k3)•
Therefore, all single-field models predict fNL≈(5/12)(1–ns).•
With the current limit ns=0.96, fNL is predicted to be 0.017.Maldacena (2003); Seery & Lidsey (2005); Creminelli & Zaldarriaga (2004)
* for which the single field is solely responsible for driving inflation and generating observed fluctuations. 19
Understanding the Theorem
•
First, the squeezed triangle correlates one very long-wavelength mode, kL (=k3), to two shorter wavelength modes, kS (=k1≈k2):
•
<ζk1ζk2ζk3> ≈ <(ζkS)2ζkL>•
Then, the question is: “why should (ζkS)2 ever care about ζkL?”•
The theorem says, “it doesn’t care, if ζk is exactly scale invariant.”20
ζ kL rescales coordinates
•
The long-wavelengthcurvature perturbation rescales the spatial
coordinates (or changes the expansion factor) within a
given Hubble patch:
•
ds2=–dt2+[a(t)]2e2ζ(dx)2ζkL
left the horizon already
Separated by more than H-1
x1=x0eζ1 x2=x0eζ2
21
ζ kL rescales coordinates
•
Now, let’s put small-scale perturbations in.•
Q. How would theconformal rescaling of coordinates change the
amplitude of the small-scale perturbation?
ζkL
left the horizon already
Separated by more than H-1
x1=x0eζ1 x2=x0eζ2 (ζkS1)2 (ζkS2)2
22
ζ kL rescales coordinates
•
Q. How would theconformal rescaling of coordinates change the
amplitude of the small-scale perturbation?
•
A. No change, if ζk is scale- invariant. In this case, nocorrelation between ζkL and (ζkS)2 would arise.
ζkL
left the horizon already
Separated by more than H-1
x1=x0eζ1 x2=x0eζ2 (ζkS1)2 (ζkS2)2
23
Real-space Proof
•
The 2-point correlation function of short-wavelength modes, ξ=<ζS(x)ζS(y)>, within a given Hubble patchcan be written in terms of its vacuum expectation value (in the absence of ζL), ξ0, as:
•
ξζL ≈ ξ0(|x–y|) + ζL [dξ0(|x–y|)/dζL]•
ξζL ≈ ξ0(|x–y|) + ζL [dξ0(|x–y|)/dln|x–y|]•
ξζL ≈ ξ0(|x–y|) + ζL (1–ns)ξ0(|x–y|)Creminelli & Zaldarriaga (2004); Cheung et al. (2008)
3-pt func. = <(ζS)2ζL> = <ξζLζL>
= (1–ns)ξ0(|x–y|)<ζL2>
•
ζS(x)•
ζS(y)24
Where was “Single-field”?
•
Where did we assume “single-field” in the proof?•
For this proof to work, it is crucial that there is onlyone dynamical degree of freedom, i.e., it is only ζL that modifies the amplitude of short-wavelength modes, and nothing else modifies it.
•
Also, ζ must be constant outside of the horizon(otherwise anything can happen afterwards). This is also the case for single-field inflation models.
25
Therefore...
•
A convincing detection of fNL > 1 would rule out all of the single-field inflation models, regardless of:•
the form of potential•
the form of kinetic term (or sound speed)•
the initial vacuum state•
A convincing detection of fNL would be a breakthrough.26
Large Non-Gaussianity from Single-field Inflation
•
S=(1/2)∫d4x √–g [R–(∂μφ)2–2V(φ)]•
2nd-order (which gives Pζ)•
S2=∫d4x ε [a3(
∂tζ)2–a(∂iζ)2]•
3rd-order (which gives Bζ)•
S3=∫d4x ε2 […a3(
∂tζ)2ζ+…a(∂iζ)2ζ +…a3(
∂tζ)3] + O(ε3)27
Cubic-order interactions are suppressed by an additional factor of ε. (Maldacena 2003)
Large Non-Gaussianity from Single-field Inflation
•
S=(1/2)∫d4x √–g {R–2P[(∂μφ)2,φ]} [general kinetic term]•
2nd-order•
S2=∫d4x ε [a3(
∂tζ)2/cs2–a(∂iζ)2]•
3rd-order•
S3=∫d4x ε2 […a3(
∂tζ)2ζ/cs2 +…a(∂iζ)2ζ +…a3(
∂tζ)3/cs2] +O(ε3)
28
Some interactions are enhanced for cs2<1.
(Seery & Lidsey 2005; Chen et al. 2007)
“Speed of sound”
cs2=P,X/(P,X+2XP,XX)
•
S=(1/2)∫d4x √–g {R–2P[(∂μφ)2,φ]} [general kinetic term]•
2nd-order•
S2=∫d4x ε [a3(
∂tζ)2/cs2–a(∂iζ)2]•
3rd-order•
S3=∫d4x ε2 […a3(
∂tζ)2ζ/cs2 +…a(∂iζ)2ζ +…a3(
∂tζ)3/cs2] +O(ε3)
Large Non-Gaussianity from Single-field Inflation
29
Some interactions are enhanced for cs2<1.
(Seery & Lidsey 2005; Chen et al. 2007)
“Speed of sound”
cs2=P,X/(P,X+2XP,XX)
Another Motivation For f NL
•
In multi-field inflation models, ζk can evolve outside the horizon.•
This evolution can give rise to non-Gaussianity;however, causality demands that the form of non-
Gaussianity must be local!
Separated by more than H-1
x1 x2
ζ(x)=ζg(x)+(3/5)fNL[ζg(x)]2+Aχg(x)+B[χg(x)]2+… 30
Now:
•
I hope that I could convince you that fNL is a very powerful quantity for testing single-field inflation models.•
Let’s look at the observational data!31
32
Decoding Bispectrum
•
Hydrodynamics at z=1090 generates acousticoscillations in the bispectrum
•
Well understood at the linear level (Komatsu &Spergel 2001)
•
Non-linear extension?•
Nitta, Komatsu, Bartolo,Matarrese & Riotto, arXiv:
0903.0894
•
fNLlocal~0.5Measurement
•
Use everybody’s favorite: χ2 minimization.•
Minimize:•
with respect to Ai=(fNLlocal, fNLequilateral, bsrc)•
Bobs is the observed bispectrum•
B(i) is the theoretical template from various predictions33
Journal on f NL (95%CL)
•
–3500 < fNL < 2000 [COBE 4yr, lmax=20 ]•
–58 < fNL < 134 [WMAP 1yr, lmax=265]•
–54 < fNL < 114 [WMAP 3yr, lmax=350]•
–9 < fNLlocal < 111 [WMAP 5yr, lmax=500]Komatsu et al. (2002) Komatsu et al. (2003)
Spergel et al. (2007) Komatsu et al. (2008)
34
Latest on f NL
•
CMB (WMAP5 + most optimal bispectrum estimator)•
–4 < fNL < 80 (95%CL)•
fNL = 38 ± 21 (68%CL)•
Large-scale Structure (Using the SDSS power spectra)•
–29 < fNL < 70 (95%CL)•
fNL = 31 +16–27 (68%CL)Smith, Senatore & Zaldarriaga (2009)
Slosar et al. (2009)
(Fast-moving field!)
35
Weak 2- σ “Hint”?
•
So, currently we have something like fNL~40±20 from the WMAP 5-year data, and 30±15 from WMAP5+LSS.•
Without a doubt, we need more data...•
WMAP7 is coming up (early next year)•
WMAP9 in ~2011–2012•
And...36
Planck!
•
Planck satellite is scheduled to be launched TOMORROW, from French Guiana.•
Planck’s expected 68%CL errorbar is ~5.•
Therefore, if fNL~40, we would see it at 8σ. If ~30, 6σ.Either way, IF (big if) fNL~30–40, we will see it
unambiguously with Planck, which is expected to deliver the first-year results in ≥2012.
37
Trispectrum: Next Frontier?
•
The local form bispectrum,Βζ(k1,k2,k3)=(2π)3δ(k1+k2+k3)fNL[(6/5)Pζ(k1)Pζ(k2)+cyc.]
•
is equivalent to having the curvature perturbation in position space, in the form of:•
ζ(x)=ζg(x) + (3/5)fNL[ζg(x)]2•
This provides a useful model to parametrize non-Gaussianity, and generate initial conditions for, e.g., N-body simulations.
•
This can be extended to higher-order:•
ζ(x)=ζg(x) + (3/5)fNL[ζg(x)]2 + (9/25)gNL[ζg(x)]3 38Local Form Trispectrum
•
For ζ(x)=ζg(x) + (3/5)fNL[ζg(x)]2 + (9/25)gNL[ζg(x)]3, we obtain the trispectrum:•
Tζ(k1,k2,k3,k4)=(2π)3δ(k1+k2+k3+k4){gNL[(54/25)Pζ(k1)Pζ(k2)Pζ(k3)+cyc.] +
(fNL)2[(18/25)Pζ(k1)Pζ(k2)(Pζ(|k1+k3|)+Pζ(|k1+k4|))+cyc.]}
k3
k4
k2
k1
g NL
k2
k1
k3
k4
f NL 2
39Trispectrum: if f
NLis ~50,
excellent cross-check for Planck
• Trispectrum (~
fNL2)
• Bispectrum (~
fNL)
Kogo & Komatsu (2006)
40
(Slightly) Generalized Trispectrum
•
Tζ(k1,k2,k3,k4)=(2π)3δ(k1+k2+k3+k4){gNL[(54/25)Pζ(k1)Pζ(k2)Pζ(k3)+cyc.]
+τNL[(18/25)Pζ(k1)Pζ(k2)(Pζ(|k1+k3|)+Pζ(|k1+k4|))+cyc.]}
The local form consistency relation, τNL=(fNL)2, may not be respected – additional test of multi-field inflation!
k3
k4
k2
k1
g NL
k2
k1
k3
k4
f NL 2
41Trispectrum: Next Frontier
•
A new phenomenon: many talks given at the IPMU non- Gaussianity workshop emphasized the importance ofthe trispectrum as a source of additional information on the physics of inflation.
•
τNL ~ fNL2; τNL ~ fNL4/3; τNL ~ (isocurv.)*fNL2; gNL ~ fNL; gNL ~ fNL2; or they are completely independent•
Shape dependence? (Squares from ghost condensate, diamonds and rectangles from multi-field, etc)42
Large-scale Structure of the Universe
•
New frontier: large-scale structure of the universe as a probe of primordial non-Gaussianity43
New, Powerful Probe of f
NL•f
NLmodifies the power spectrum of galaxies on very large scales
–Dalal et al.; Matarrese & Verde –Mcdonald; Afshordi & Tolley
•The statistical power of this method is VERY promising
–SDSS: –29 < fNL < 70 (95%CL);
Slosar et al.
–Comparable to the WMAP 5-year limit already
–Expected to beat CMB, and reach a
sacred region: fNL~1 44
Effects of f NL on the statistics of PEAKS
•
The effects of fNL on the power spectrum of peaks (i.e., galaxies) are profound.•
How about the bispectrum of galaxies?45
Previous Calculation
•
Scoccimarro, Sefusatti & Zaldarriaga (2004); Sefusatti &Komatsu (2007)
•
Treated the distribution of galaxies as a continuousdistribution, biased relative to the matter distribution:
•
δg = b1δm + (b2/2)(δm)2 + ...•
Then, the calculation is straightforward. Schematically:•
<δg3> = (b1)3<δm3> + (b12b2)<δm4> + ...Non-linear Bias Bispectrum Non-linear Gravity
Primordial NG 46
Previous Calculation
•
We find that this formula captures only a part of the full contributions. In fact, this formula is sub-dominant in the squeezed configuration, and the new terms are dominant.Non-linear Bias
Non-linear Gravity
Primordial NG
47
Non-linear Gravity
•
For a given k1, vary k2 and k3, with k3≤k2≤k1•
F2(k2,k3) vanishes in the squeezed limit, and peaks at theelongated triangles. 49
Non-linear Galaxy Bias
•
There is no F2: less suppression at the squeezed, and less enhancement along the elongated triangles.•
Still peaks at the equilateral or elongated forms. 50Primordial NG (SK07)
•
Notice the factors of k2 in the denominator.•
This gives the peaks at the squeezed configurations. 51New Terms
•
But, it turns our that Sefusatti & Komatsu’s calculation, which is valid only for the continuous field, misses the dominant terms that come from the statistics ofPEAKS.
•
Jeong & Komatsu, arXiv:0904.0497Donghui Jeong 52
MLB Formula
•
N-point correlation function of peaks is the sum of M- point correlation functions, where M≥N.Matarrese, Lucchin & Bonometto (1986)
53
Bottom Line
•
The bottom line is:•
The power spectrum (2-pt function) of peaks issensitive to the power spectrum of the underlying mass distribution, and the bispectrum, and the trispectrum,
etc.
•
Truncate the sum at the bispectrum: sensitivity to fNL•
Dalal et al.; Matarrese&Verde; Slosar et al.;Afshordi&Tolley
54
Bottom Line
•
The bottom line is:•
The bispectrum (3-pt function) of peaks is sensitive to the bispectrum of the underlying mass distribution, and the trispectrum, and the quadspectrum, etc.•
Truncate the sum at the trispectrum: sensitivity to τNL (~fNL2) and gNL!•
This is the new effect that was missing in Sefusatti &Komatsu (2007).
55
Real-space 3pt Function
•
Plus 5-pt functions, etc... 56New Bispectrum Formula
•
First: bispectrum of the underlying mass distribution.•
Second: non-linear bias•
Third: trispectrum of the underlying mass distribution.57Local Form Trispectrum
•
For general multi-field models, fNL2 can be more generic: often called τNL.•
Exciting possibility for testing more about inflation! 58Φ =(3/5) ζ
Local Form Trispectrum
k3
k4
k2
k1
g NL
k2
k1
k3
k4
f NL 2 (or τ NL )
59
Φ =(3/5) ζ
60
Shape Results
•
The primordial non-Gaussianity terms peak at the squeezed triangle.•
fNL and gNL terms have the same shape dependence:•
For k1=k2=αk3, (fNL term)~α and (gNL term)~α•
fNL2 (τNL) is more sharply peaked at the squeezed:•
(fNL2 term)~α361
Key Question
•
Are gNL or τNL terms important?62
63
1/k2
64
Summary
•
Non-Gaussianity is a new, powerful probe of physics of the early universe•
It has a best chance of ruling out all of the single-field inflation models at once.•
fNL ~ 2σ at the moment, wait for WMAP 9-year (2011) and Planck (≥2012) for more σ’s (if it’s there!)•
To convince ourselves of detection, we need to see theacoustic oscillations, and the same signal in the bispectrum and trispectrum, of both CMB and the large-scale structure
of the universe. 65
Now, let’s pray:
•
May Planck succeed!66
Now, let’s pray:
• May the signal be there!
67